The paper studies a mathematical model called a Fleming-Viot particle system, where a population of particles moves around a network of locations and occasionally gets "killed." When a particle is killed, it is immediately replaced by a copy of one of the surviving particles. Over time, you can track which particles are ancestral to the current population, and the paper asks whether that ancestral lineage tends to flow in a preferred direction around the network, or whether it is perfectly balanced with no net circulation.
The central finding is that two particles is the precise dividing line for this kind of directional bias. With exactly two particles, the ancestry never develops a preferred direction regardless of how the network is set up or how likely killing is at each location. This is a clean, universal result. With three or more particles, however, directional circulation in the ancestral flow generically appears, even when the underlying movement of particles is perfectly reversible (meaning it has no inherent preferred direction on its own). The paper shows this rigorously on a three-location network, demonstrating that for almost every possible assignment of killing rates, the ancestral current flows more strongly one way around the network than the other.
The paper also gives a concrete probabilistic interpretation of what this circulation means in biological or ecological terms. When migration events between locations are rare, the mathematical circulation translates into an asymmetry in evolutionary outcomes: a lineage is more likely to spread from one location to another and take over the population in one direction of travel than in the reverse direction. This is surprising because the raw movement rules are symmetric, yet the combination of copying upon death and population-level competition breaks that symmetry and produces a directional bias in which evolutionary paths are most likely.